Moral Hazard Misconceptions: the Case of the Greenspan Put

Moral Hazard Misconceptions: the Case of the Greenspan Put Gideon Bornstein Northwestern University Guido Lorenzoni Northwestern University December 2017 Abstract Policy discussions on financial regulation tend to assume that whenever a corrective policy is used ex post to ameliorate the effects of a crisis, there are negative side effects in terms of moral hazard ex ante. This paper shows that this is not a general theoretical prediction, focusing on the case of monetary policy interventions ex post. In particular, we show that if the central bank does not intervene by monetary easing following a crisis, an aggregate demand externality makes borrowing ex ante inefficient. If instead the central bank follows the optimal discretionary policy and intervenes to stabilize asset prices and real activity, we show examples in which the aggregate demand externality disappears, reducing the need for ex ante intervention. For very useful comments we thank Eduardo Davila, Olivier Jeanne, the editor, and two referees. 1

1 Introduction Many economists and commentators have remarked that the conduct of interest policy by the central bank can affect the incentives of the financial sector. In particular, a common complaint is that the so called Greenspan put encourages excessive leverage and risk taking by banks. The argument goes as follows. Suppose that when an adverse shock drives down asset prices, the central bank intervenes systematically by lowering interest rates. This becomes an implicit commitment to prop up asset prices in times of distress and encourages banks and other financial players to borrow more and take on more risk ex ante. In turn, this increases the risk of systemic crises, with possible harmful repercussions on aggregate activity. Therefore, countercyclical interest rate policy may end up increasing, rather than reducing, macroeconomic volatility. Countercyclical monetary policy generates a form of moral hazard, where financial firms do not receive help directly, but are subsidized indirectly by the central bank s low interest rate policy. A number of recent papers have formalized this idea in different ways, including Lorenzoni 2001), Diamond and Rajan 2012), Farhi and Tirole 2012), and Chari and Kehoe 2016). This paper attacks the problem from a different perspective. While the existing literature emphasizes the distortionary role of an overly active monetary policy, here we emphasize the distortions that arise when monetary policy is too passive. In other words, we show that the lack of countercyclical interventions, in the face of negative aggregate shocks, can also worsen the problem of excessive leverage in the financial sector. While the existing literature emphasizes a form of pecuniary externality, which travels through asset prices, this paper emphasizes the presence of an aggregate demand externality, which travels through the level of real output. In this respect, our model builds on the recent literature on macroprudential policy and aggregate demand externalities started by Farhi and Werning 2016), Korinek and Simsek 2016), and Schmitt-Grohé and Uribe 2106). To make our main point, we present a simple model in which a countercyclical interest rate policy can completely eliminate the overborrowing distortion. We first consider a monetary policy regime in which the central bank keeps the interest rate constant and does not respond at all to financial shocks that move asset prices and real activity. In that regime, the equilibrium is constrained inefficient and there is overborrowing. We then consider a regime in which the central bank chooses the interest rate optimally under discretion. In that regime monetary policy is countercyclical and completely eliminates both output volatility and asset price volatility. Such a policy does indeed induce higher borrowing ex ante. However, under this policy the overborrowing problem disappears. 2

So even though the level of borrowing goes up, the distance between the laissez-faire level of borrowing and the socially constrained) efficient level goes down. In this economy, ex post monetary interventions and ex ante regulation are substitutes from a social welfare point of view: a more active monetary intervention ex post decreases the need for regulation to mitigate overborrowing ex ante. The result above arises under special assumptions regarding risk aversion and the nature of the shocks hitting the economy. So we also investigate other parameter configurations and other policy regimes to better disentangle the forces that make ex post monetary policy and ex ante regulation complements or substitutes. A common feature of all our examples is that complementarity between monetary policy and regulation only arises if the monetary authority goes over and above a simple objective of reducing the output gap. Two closely related papers are Benigno et al. 2013) and Korinek and Jeanne 2016), who analyze the relative benefits of ex ante and ex post policies to deal with financial instability in different models. Korinek and Jeanne 2016) look at a model in which the ex post policy is a form of bailout in which resources are transferred from lenders to borrowers. Proposition 6 in their paper shows that when more resources are available for an ex post bailout, the need for ex ante macroprudential policy may be reduced. That is, ex ante and ex post policies can be substitutes rather than complements. That result is close in spirit to the message of this paper. However, their model is a purely real model with an exogenous interest rate pinned down by a storage technology, while here we focus on the use of interest rate policy as a tool to stabilize asset prices and deal with a crisis ex post. Benigno et al. 2013) is closer to our model, in that the ex post intervention is also captured by monetary policy. 1 There are two main differences with our paper. First, our paper features an explicit role for asset prices, and thus can be used to discuss asset price stabilization by the central bank. Second, our paper focuses on a simple model where analytic results can be derived. In particular, we focus on understanding whether monetary tools and macroprudential tools are substitutes or complements, provide numerical examples of both cases, and show how the interplay of pecuniary externalities and aggregate demand externalities leads to one case or the other. An additional novelty of this paper, is the extension of the analysis to unconventional monetary policy in Section 6. Finally, our paper is related to a classic debate on the benefits of asset price stabilization as a monetary policy objective. Bernanke and Gertler 1999) made the point that monetary policy should respond to asset price movements insofar as they affect aggregate 1 See also Benigno et al. 2016) in an open economy context. 3

demand. In this paper, we look precisely at monetary responses to asset prices that are motivated by aggregate demand management and ask whether these responses encourage instability and excess leverage ex ante. We then show that these ex ante considerations don t necessarily undermine the argument of Bernanke and Gertler 1999), and may in fact strengthen it. Monetary policy that aims to reduce the output gap can make the economy more efficient also by reducing the externalities associated with ex ante financial decisions. The paper is organized as follows. Section 2 describes the model. Section 3 characterizes equilibria under different monetary regimes. Section 4 contains the main result on the positive and normative implications of countercyclical monetary policy. Section 5 contains a variety of other examples. Section 6 analyzes a model where the zero lower bound is binding and monetary policy intervenes by quantitative easing. Section 7 concludes. 2 The model Consider a three period economy, with t = 0, 1, 2. In periods 0 and 2 the economy is an endowment economy. In period 1 output is produced with a linear technology that uses only labor. There are two groups of agents of equal size, labeled A and B. The preferences of A agents are represented by the utility function ) ) E c0 A + uca 1 ) v n1 A + u c2 A, 1) where ct A denotes consumption, n1 A v have the isoelastic forms denotes labor effort in period 1. The functions u and u c) = 1 1 γ c1 γ, v n) = ψ 1 + φ n1+φ. The preferences of B agents are represented by the utility function ) ) E c0 B + βu c1 B + β 2 u c2 B, 2) where ct B denotes consumption and 0 < β < 1. The assumption of linear utility for both agents in period 0 simplifies the welfare analysis, by making utility transferable ex ante. Both groups of agents receive a large endowment of the single consumption good in period 0. In period 1, A agents receives labor income by supplying labor on a competitive 4

labor market at the wage rate w 1. There is a discrete set of states of the world s S, with probability distribution π s). The state s is revealed in period 1. There is a risky asset in fixed supply in the economy, which pays δ 1 s) units of consumption good in period 1 and δ 2 s) units in period 2. The risky asset can only be held by B agents. In periods 0 and 1, agents trade a real non-state contingent bond that pays 1 unit of consumption goods in the following period. Assuming that only B agents can hold the risky asset and that they are more impatient than A agents is a simple way of obtaining levered agents, whose balance sheets are exposed to shocks and affect aggregate spending in the economy. The channels captured in this simple model can be extended to richer models of borrowing and lending as, for example, models with levered intermediaries exposed to aggregate risk who make lending decisions that affect aggregate spending. In period 1, there is a continuum of monopolistic firms on the interval 0, 1 that produce intermediate goods. Each intermediate good is produced with a linear technology that uses only labor x j = n j1. The goods are then combined to produce the final consumption good using the production function 1 Y = 0 ɛ x ɛ 1 ɛ j1 dj ɛ 1, with ɛ > 1. The firms selling the differentiated goods set the price of their good p j1 one period in advance, in period 0. The firms are fully owned by A agents. The aggregate nominal price level in period t is denoted by p t. 2 Each firm j sets its price to maximize nominal expected profits discounted using the nominal stochastic discount factor of A agents u c A 1 )/p 1. That is, firm j maximizes u c1 A E ) ) 0 1 + σ)pj1 x p j1 w 1 x j1 1 taking as given the demand function for good j, which is x j1 = ) ɛ p j1 /p 1 Y. In the expression above, w 1 is the nominal wage in period 1 and σ is a proportional subsidy to the production of intermediate goods. We assume the government finances the subsidy σ by levying a lump-sum tax on A agents, so that the real income of A agents in period 1 is simply Y. We introduce the subsidy to give the government a tool to correct for the monopolistic distortion. 2 The aggregate price in the economy in period 1 is equal to 1 0 p1 ɛ j1 ) 1 dj 1 ɛ. 5

Policy is captured by three instruments: the nominal interest rate i t, the subsidy σ, and a macroprudential linear tax τ on borrowing at t = 0. The tax revenues from the macroprudential tax are rebated as a lump sum transfer to the B agents, who pay the tax. Our assumptions on how the subsidy σ and the tax τ are financed imply that all interventions are wealth neutral, so we do not allow the government to directly reallocate resources across agents. 3 Equilibrium under different monetary regimes Our goal is to study the interaction between monetary policy and macroprudential policy. In this section, we start by characterizing the equilibrium under three different monetary policy regimes, setting the macroprudential tax to zero. In the next sections, we investigate the benefits of adding macroprudential policy under different monetary regimes. 3.1 Preliminary steps Before introducing the different regimes, we need a few preliminary steps that characterize equilibrium allocations independently of the regime. All price setters face the same strictly concave problem, so they all choose the same nominal price p j1 = p 1 and they all produce the same amount of goods Y with the same amount of labor n = Y. Since prices are set one period in advance, monetary policy can determine the real interest rate between periods 1 and 2, which we denote by r 1 + i 1 ) p 1 p 2 1. We can then focus on characterizing the equilibrium allocations that can be achieved by a social planner who chooses r at date 1 and the subsidy σ at date 0. To characterize the equilibrium, we go backwards in time and solve first for the equilibrium at dates t = 1, 2. In this way, we obtain the output level Y that arises in period 1, in state s, when A agents hold D units of real debt issued by B agents and the real interest rate is r. We use the function YD, r, s) to denote this output level and leave its explicit derivation to Appendix 8.1. We now introduce two value functions which will be useful in the rest of the analysis. Let V A b, D, r, s) denote the expected utility in state s of an agent A who enters period 1 with b units of bonds in an economy in which all other A agents hold D units of bonds, all other B agents have D units of debt, and the real interest rate is r. Let V B d, D, r, s) 6

denote the analogous value for the B agents, with d denoting the individual level of debt. In equilibrium we have b = d = D, but for the analysis it is convenient to separate individual decisions from aggregates. The formal definitions of V A and V B are: V A b, D, r, s) = max c A 1,c A 2 V B d, D, r, s) = max c B 1,c B 2 uc A 1 ) v YD, r, s)) + uca 2 ) s.t. c1 A + 1 1 + r ca 2 uc1 B) + βucb 2 ) = YD, r, s) + b, s.t. c B 1 + 1 1 + r cb 2 = δ 1 s) + 1 1 + r δ 2s) d. The equations above show that the expected utility of the two agents depend on three key variables: the individual financial positions, the aggregate level of debt in the economy, and the real interest rate. The following lemma characterizes these relations. The proof is in the appendix. Lemma 1. At b = d = D the partial derivatives of the value functions are V A b b, D, r, s) = u c A 1 ), Vd Bd, D, r, s) = u c1 B), ) VD u Ab, D, r, s) = Y c1 A) v Y) D, VD B d, D, r, s) = 0, ) Vr A b, D, r, s) = u 1 ) c1 Y A) + D c1 A 1 + r + u Y c1 A) v Y) r, ) Vr B d, D, r, s) = u 1 c1 Y B) + D c1 A 1 + r. The values of c1 A and cb 1 in the expressions above are the values that solve the problems in the value functions of A and B, at b = d = D; Y is short for YD, r, s). Let us provide intuition for the results in the lemma. The first two partial derivatives in Lemma 1 are completely standard and capture the private marginal benefit of lending for agent A and the private marginal cost of borrowing for agent B. The derivative VD A captures an aggregate demand externality that plays a crucial role in the rest of the analysis. The effect of debt on output is captured by the derivative Y/ D. This derivative is negative because a debt payment is a transfer of resources from B agents to A agents, and the former have a higher propensity to consume, due to 7

higher impatience. 3 In this way, the model captures in a nutshell the real effects of debt overhang. A marginal change in output increases or decreases welfare depending on the sign of u c A 1 ) v Y). With flexible prices after removing the monopolistic distortion this difference would always be zero. But in a new Keynesian environment this term can be non zero. In particular, when it is positive, increasing output and hours worked yields marginal benefits greater than marginal costs. This is the efficiency loss typically associated to the notion of output gap. We call u c A 1 ) v Y) the labor wedge and use interchangeably the language the labor wedge is positive/negative and the output gap is positive/negative. 4 When there is a positive output gap, a higher level of debt is welfare reducing as it reduces output at a point where output increases would be welfare improving. The next derivative, VD B, is zero because the B agents decision problem is not directly affected by the level of Y. Finally, the derivatives Vr A and Vr B describe how agents utilities depend on the real interest rate. The real interest rate affects agents in two ways. First, increasing the interest rate reallocates resources from borrowers to lenders, in proportion to the amount of borrowing and lending taking place between periods 1 and 2. Since in equilibrium A agents save Y + D c1 A and this is equal to the amount borrowed by B agents, we obtain the first term in Vr A and the only term in Vr B. The real interest rate also affects A agents through the aggregate demand channel. In particular, raising the interest rate changes the equilibrium level of output according to Y/ r and this affects welfare in proportion to the labor wedge. To complete the equilibrium characterization, we need three additional steps: determine how monetary policy chooses r in each state of the world; determine the level of D which arises in equilibrium at t = 0; ensure that the price setters optimality condition is satisfied at t = 0. The solution of these steps depends on the monetary policy regime. In the next section we consider three possible regimes and complete the equilibrium characterization. 3 The formal proof that Y/ D < 0 is in the appendix. 4 Here we are thinking of the welfare benefits of producing more goods and assigning them to A agents. Given that there are two groups of agents and imperfect insurance, the notion would be different if we looked at the welfare benefits of assigning the extra goods to B agents. This would change slightly the decomposition and interpretation of our two main forces, but it would change nothing in the substance of the analysis. 8

3.2 Inertial regime To define our first two regimes, we assume that policy is set by a planner who maximizes the social welfare function W = E V A D, D, r, s) + βv B D, D, r, s), 3) with side transfers taking place at date 0 to reallocate utility between A and B agents. 5 In the first regime, the interest rate at t = 1 is set by the planner before the realization of the state s. We think of this regime as capturing inertia in the response of the central bank to information about aggregate shocks, so we call it the inertial regime. 6 The timing is as follows. First, the equilibrium debt level d = b = D is determined in the bond market at date 0. 7 Next, at the beginning of period 1, before the realization of the state s, the planner chooses a real interest rate r to maximize 3). After the realization of s, the central bank cannot revisit its choice of the interest rate. Agents choose their spending in period 1 and that determines output. Finally, in period 2, agents consume their endowment net of their final bonds positions. The subsidy σ is set to ensure that the price setting condition of the producers at t = 0 is satisfied. 8 The crucial restriction in this regime is that the central bank cannot use the interest rate policy to mitigate the drop in the asset price for low realizations of δ. As we shall see, this sluggish response of the central bank may reduce the incentive of B agents to borrow in period 0 as they cannot rely on the implicit bailout coming from looser policy. An equilibrium under this regime is given by a pair D, r that satisfies the following 5 The presence of heterogeneous discount factors introduces a source of time inconsistency in optimal policy. By focusing on the welfare function 3) we leave aside that source of time inconsistency. Other interesting sources are present, as we shall see. 6 An alternative source of real interest rate rigidity that would lead to related results would be to consider an open economy with a fixed exchange rate regime or a monetary union. 7 Since prices are flexible in period 0 the central bank can choose the nominal interest rate i 0 but has no power to affect the real interest rate between dates 0 and 1, 1 + i 0 )p 0 /p 1, which is determined at the level that clears the bond market. 8 Using the optimality condition for labor supply we obtain the equilibrium real wage w 1 p 1 u c A 1 ) = v Y). Substituting in the optimality condition of price setters we obtain E Y 1 + σ) u c1 A) ɛ ) ɛ 1 v Y) = 0. The value of σ is chosen to satisfy this condition. 9

two conditions: i) optimal monetary policy, characterized by the first-order condition E Vr A D, D, r, s) + βvr B D, D, r, s) = 0; 4) ii) bond market equilibrium at date 0, characterized by the condition E Vb A D, D, r, s) + βe V d B D, D, r, s) = 0. 5) Since the central bank chooses the interest rate r after the bonds market in period 0 has cleared, the central bank takes D as given when choosing r. At the same time, consumers do not internalize the effect that the equilibrium level of D has on the central bank s choice of r, simply because they are atomistic. This explains why strategic considerations do not appear in conditions 4) and 5). Using Lemma 1, condition 4) can be rewritten as ) E u Y c1 A) v Y) r ) ) + E u 1 c1 A) βu c1 Y B) + D c1 A = 0. 6) 1 + r This condition shows that the central bank is balancing two effects when choosing the interest rate. The first is a standard new Keynesian effect: changing the interest rate affects equilibrium output and this increases or decreases welfare depending on the sign of the output gap. The second effect is a pecuniary externality associated to incomplete markets: changing the interest rate reallocates resources from borrowers to lenders and if the marginal utilities of borrowers and lenders are different ex post this can have welfare benefits. Davila and Korinek 2017) call this a distributive externality. Here, given that r is not allowed to be state contingent, the monetary authority chooses a level of the interest rate that balances these two welfare effects in expectation. Using Lemma 1 we can also rewrite 5) as follows Eu c1 A) = βeu c1 B ), 7) which shows that the debt level is chosen to equalize expected marginal utilities. Conditions 6) and 7) will be used to characterize the equilibrium in the inertial regime in the rest of the paper. 10

3.3 Proactive regime In our second regime, the planner also aims to maximize the social welfare function 3), but can now choose a fully state contingent interest rate. The timing is as in the previous regime, but r is set after observing s. We call this regime the proactive regime. An equilibrium under this regime is given by a debt level D and interest rates {r s)} s S that satisfy the following two conditions: i) optimal monetary policy, characterized by the first-order condition V A r D, D, r s), s) + βv B r D, D, r s), s) = 0 for all s S; 8) ii) bond market equilibrium at date 0, characterized by the condition E Vb A D, D, r s), s) + βe V d B D, D, r s), s) = 0. 9) As in the previous regime, Lemma 1 can be used to rewrite these conditions. In particular, condition 6) now holds state by state and takes the form ) u Y ) ) c1 A) v Y) r + u 1 c1 A) βu c1 Y B) + D c1 A 1 + r Condition 9) can be written as 7), as in the inertial regime. = 0. 10) As in the previous regime, the subsidy σ is chosen to ensure that price-setters optimality is satisfied. 3.4 Output targeting regime The third regime is one in which the planner responds to the state s, but its only objective is to replicate the equilibrium that would arise under flexible prices and perfect competition i.e., with no monopolistic distortion). So we are looking at a regime in which σ and rs) are set so that the following condition holds state by state u c A 1 ) v Y) = 0. We call this regime the output targeting regime, because it keeps the output gap at zero. The ex ante optimality condition that ensures equilibrium in the debt market at t = 0 is 7) as in the previous regimes. The conditions just described, together with the conditions characterizing a continuation equilibrium, are sufficient to pin down the equilibrium allocation under this regime. 11

However, for the analysis to follow we need to be precise on the timing. We assume that the central bank commits at t = 0 to set at t = 1 the real interest rate r s) that implements the flexible price allocation. Therefore this regime is different from the previous two not only because of the planner s objective, but also for its commitment to future interest rates. As we shall see, this commitment assumption will deliver a clean characterization of this regime in terms of macroprudential policy. 4 More borrowing, less overborrowing We now turn to our main questions: How does the monetary regime affect the level of borrowing? How does it affect the benefits of macroprudential intervention? In a simple example with log preferences, the answers to these two questions go in opposite directions: the proactive regime yields more debt than the inertial regime, but, at the same time, it reduces the benefits of macroprudential policy. In fact, it makes macroprudential policy completely unnecessary. This example makes the main point of the paper. In this section, we present the log example and provide some intuition for it. In the next section, we go beyond the example to investigate more generally the implications of our analysis. Let us specialize our model by assuming log preferences, u c) = log c, and no dividend at t = 1, δ 1 s) = 0. Consumption levels at t = 1 are now c1 A = 1 Y + D), 2 and Adding them up and solving for Y gives aggregate output c1 B = 1 ) δ2 1 + β 1 + r D. 11) Y = 2 δ 2 1 + β 1 + r 1 β 1 + β D. The following proposition characterizes the equilibrium in the proactive regime, which, in this case, is identical to the output targeting regime. The proof is in the appendix. Proposition 1. With log preferences and δ 1 = 0, the proactive regime and the output targeting regime coincide and they are characterized by a real interest rate that exactly offsets changes in δ 2, so the asset price δ 2 /1 + r), output, and the consumption levels of both agents are constant across states. 12

In this example, optimal discretionary monetary policy achieves perfect stabilization of asset prices through interest rate policy an especially stark example of the Greenspan put at work. This policy is optimal because it achieves, at the same time, a zero output gap and perfect insurance between agents, thus making both terms of 10) zero. The insurance effect of the policy can be seen immediately from 11), that shows that the wealth of B agents is perfectly stabilized by keeping δ 2 /1 + r) constant. Turning to the inertial regime, it is easy to see that we no longer obtain perfect insurance for B agents, as now the interest rate is constant and asset prices move with δ 2, making consumption more volatile. We can then ask whether this increased volatility induces B agents to be more cautious and borrow less ex ante. Substituting the consumption values just derived, equation 7) takes the form 2 E Y + D = βe 1 + β δ 2 / 1 + r) D Since the utility function displays prudence, this equation suggests that higher volatility of the asset price δ 2 /1 + r) leads to increased savings, that is, less borrowing, by B agents. This intuition is confirmed in the next proposition. To prove the proposition requires some additional steps, because in the inertial regime the variable Y is also volatile, and not only the volatility, but also the average levels of r and Y are different in the two regimes. To deal with these complications, we restrict attention to the case of small shocks to δ 2. The proof is in the appendix. Proposition 2. With log preferences, δ 1 = 0, and small shocks to δ 2, the proactive regime features a higher level of borrowing D in equilibrium than the inertial regime. This result seems to offer a perfect example of the evil incentive effects of the Greenspan put. However, this is just a positive result about the levels of D in the two regimes, not about their efficiency properties. The next proposition shows that looking at efficiency leads to different conclusions. To evaluate the benefits of macroprudential policy, we look at the effect of a marginal change in D on welfare. Proposition 3. With log preferences and δ 1 = 0, in the inertial regime there is excessive borrowing ex ante, that is, social welfare is locally decreasing in D: dw dd =. d dd E V A D, D, r, s) + βv B D, D, r, s) < 0. 13

In the proactive regime the level of borrowing is socially efficient: dw dd = d dd E V A D, D, r s), s) + βv B D, D, r s), s) = 0. In order to build intuition for this result, let us first provide some derivations that characterize the marginal social benefit of a change in D. Using the results in Lemma 1 and the private optimality condition for debt 7), we can characterize the marginal effect of debt on social welfare as follows. Lemma 2. In all the monetary regimes considered, the marginal welfare effects of a change in D is equal to dw ) dd = E u Y c1 A) v Y). 12) D This result is independent of the functional form assumptions made for u and v, and it does not rely on the special assumptions made in this section. Furthermore, it holds in all the monetary regimes considered. One may wonder why there is no term capturing the fact that changing D ex ante will affect the choice of r ex post. In the first two regimes, where monetary policy is chosen under discretion, this happens because the ex post choice of r is optimal, so an envelope argument implies that this effect can be ignored. In the output targeting regime, this happens because the central bank commits to r ex ante and does not change it off the equilibrium path if D is changed. This is where the assumption of commitment in the third regime helps to simplify the analysis. The effect of debt on output is given by Y D = 1 β 1 + β < 0, which is constant and independent of the state of the world. So we can focus on establishing the sign of the average labor wedge E u c1 A) v Y). In the inertial regime, it is possible to show that the output gap is positive in low δ 2 states and negative in high δ 2 states. This is due to the fact that in that regime the risk δ 2 is not insured and a reduction in δ 2 reduces the wealth of the B agents who have a relatively larger marginal propensity to consume. Therefore the sign of E u c A 1 ) v Y) depends on whether the positive wedge in the bad states dominates the negative wedge in the good states. In economic terms, a larger debt stock D worsens recessions in bad 14

states, which is bad for welfare, and dampens booms in good states, which is good for welfare. The proof of Proposition 3 in the appendix shows that the first effect dominates, so, overall, more borrowing reduces welfare. In the proactive regime, we have already established that the output gap is always zero. Therefore, there is no welfare gain from changing the level of borrowing ex ante, as the level of output is already at its socially efficient level state by state. Summing up, using interest rate policy to fight a recession leads to higher borrowing ex ante, but this is not a symptom of inefficient borrowing ex ante. In fact, the presence of optimal discretionary monetary policy ex post eliminates the distance between equilibrium borrowing and its socially efficient level, thus making macroprudential policy unnecessary. It is useful to connect our results here with results in Farhi and Werning 2016) and Korinek and Simsek 2016) on macroprudential policy at the zero lower bound. So far we have ignored the possibility of the zero lower bound binding, which is fine in our context because the economy has flexible prices at t = 2 so the monetary authority can choose an inflation rate between t = 1 and t = 2 to achieve any real interest rate desired. However, it is easy to add an ad hoc form of ZLB by imposing r r. Consider now a central bank that tries to achieve a zero output gap but is constrained by the ZLB. Then, the output gap can never be positive, as increasing rates is always possible. Therefore, if the ZLB is binding with positive probability the expression E u c1 A) v n1 A ) is positive and the aggregate demand externality yields excessive borrowing in line with results in Farhi and Werning 2016) and Korinek and Simsek 2016). Our results show that ex post monetary policy inertia provides a rationale for macroprudential policy even if the ZLB is not binding. We will return to the ZLB in Section 6. 5 Tradeoffs, complementarity and substitutability In the previous section we established that, under certain assumptions, ex post monetary intervention is a substitute for macroprudential policy: more intervention ex post reduces the marginal benefit of macroprudential policy. In this section we move beyond the special assumptions of the previous section and ask more generally in what cases is ex post monetary intervention a substitute for macroprudential policy and in what cases there can be complementarity between the two policies. In the analysis of the previous section, the proactive regime achieves the same allocation as the output targeting regime. As argued above, this happens because the proactive regime is able to set both terms of 10) to zero at the same time. By choosing the state 15

contingent interest rate r the central bank can take care of both frictions present in our environment: price rigidity and lack of insurance against δ shocks. This is a knife-edge result that only holds with log utility. With log utility, changing the interest rate has no direct effect on the consumption decisions of A agents, as income and substitution effects cancel each other. For B agents, the interest rate affects their consumption decisions only by changing the discounted value of dividends in period 2. So, by moving interest rates proportionally to dividends, the central bank can reach constant asset prices and constant ouptut without distorting the agents intertemporal margin. This is no longer true when income and substitutions effects do not cancel each other, i.e., when the elasticity of intertemporal substitution is different than 1. Once we move away from log utility, the central bank faces a tradeoff when setting interest rates and needs to balance a traditional macro objective stabilizing aggregate output and a financial stability objective using the interest rate to redistribute in favor of B agents hit by a negative shock. In this section we want to investigate the role of macroprudential policy when monetary policy has to deal with this tradeoff. We will make our main points using numerical examples. Our main example uses the parameters β = 0.5, γ = 0.5, ψ = 1, φ = 1, δ 1 = 0, and a binary distribution for the δ 2 shock with values 1.05 and 1.5 that occur with equal probabilities 1/2. The reason why we assume γ < 1, is that in this simple setup this assumption ensures that in the output targeting regime or, equivalently, under flexible prices, asset prices and the consumption of B agents are monotone increasing in δ 2. The reason why we choose a low discount factor for B agents is to give them a strong incentive to borrow, so we can better visualize the potential inefficiencies associated with overborrowing. In the appendix, we show that the qualitative results identified here also hold in examples with higher β. 5.1 Output stability and financial stability First, let us understand better the nature of the tradeoff faced by the monetary authority ex post. To do so, it is useful to concentrate on the equilibrium in the output targeting regime, where, by definition, the first term in 10) is set to zero. Equilibrium in the bonds market at t = 1 requires E u c1 A) βu c1 B) = 0. 16

Table 1: Interest rates in the output targeting regime and in the proactive regime Output gap targeting Proactive regime δ 2 = 1.05 6.04% 4.33% δ 2 = 1.5 29.74% 32.14% As long as c1 A and cb 1 δ 2, we have are, respectively, monotone decreasing and monotone increasing in u c A 1 ) βu c B 1 ) < 0 in the low δ 2 states and u c A 1 ) βu c B 1 ) > 0 in the high δ 2 states. This means that, starting from pure output gap stabilization, the central bank s incentive is to further lower interest rates in the bad state and to further increase them in the good state. That is, the central bank is driven to stabilize asset prices over and above what is required to stabilize output. This happens because of the insurance motive: moving one dollar from lenders to borrowers in the bad state means transferring it to agents with higher marginal utility. Reducing the interest rate achieves this transfer. In Table 1 we show how the interest rate r responds to δ 2 in the output targeting regime and in the proactive regime. In the second case, the central bank uses the interest rate more aggressively in response to bad shocks. Now we are in a situation in which the central bank is tempted to ease more in a recession caused by weak borrowers balance sheets than what it would be warranted under a pure output stabilization objective. In this situation, we ask again whether macroprudential policy may be of help. 5.2 An example of complementarity As we did above, let us first focus on the marginal benefit of changing D starting from no borrowing tax. Equation 12) continues to hold in this case. In the output targeting regime it gives dw/dd = 0. So in that regime a macroprudential tax gives no welfare benefit. In the proactive regime, on the other hand, this expression is in general different from zero once γ =1. In the numerical example above, the value of this expression is negative. Turning to optimal policy, we illustrate graphically the optimal choice of macroprudential and monetary policy in Figure 1, where we plot the contours of the social welfare 17

3 2 Borrowing tax %) 1 0-1 Proactive regime Output targeting regime -2-3 4 4.5 5 5.5 6 Interest rate in the bad state %) Figure 1: Monetary policy, macroprudential tax and welfare: complementarity function in the space spanned by the borrowing tax τ and the interest rate in the bad state, denoted by r B. To represent the policy space in two dimensions, for each point r B, τ) we set the interest rate in the good state, denoted by r G, as a linear function r G = α 0 + α 1 r B, choosing α 0, α 1 so that we go through the pairs r B, r G corresponding to the output targeting regime and to the proactive regime with an optimal macroprudential tax, depicted by the two red points in the figure. In the example considered here, monetary policy and macroprudential policy are complements, because, along the line that connects the two regimes considered, the cross derivative 2 W/ r B τ is negative. 9 That is, if we compare the output targeting regime to the proactive regime, the choice of a lower rate in the second regime increases the marginal benefit of the borrowing tax. This leads to an optimal tax in the proactive regime that is larger than the optimal tax in the output targeting regime which is zero). We derive two conclusions from this analysis. One is that once we move away from the knife-edge of Section 4, optimal policy requires a combination of monetary tools and macroprudential tools. So monetary policy is no longer a perfect substitute for macroprudential policy. The second conclusion is that we finally have a result more in line 9 Complementarity/substitutability is defined in this way in this policy space, because lowering the rate more in the bad state is a more aggressive intervention. 18

with conventional wisdom on moral hazard effects. Comparing a regime of pure output targeting to a fully discretionary regime, in the latter regime the central bank increases its interest rate response, stabilizes asset prices more, and there is a stronger motive to impose a borrowing tax ex ante. Digging a bit more in the intuition, shows that this result relies on the central bank exceeding its output objectives. In the bad state of the world the central bank is choosing to overstimulate the economy, so we are reaching an allocation with u c1 A) v n1 A ) < 0. In the good state of the world the opposite is true. Inducing agents to borrow less ex ante leads to more output ex post. This is bad for welfare in the bad state of the world where the economy is already overstimulated. But it is good in the good state of the world in which output is below potential. The latter effect dominates in our example, so overall a borrowing tax is welfare improving. This explains why there is complementarity between monetary policy and macroprudential policy in this example. 5.3 An example of substitutability In Figure 1 we compared the proactive regime and the output targeting regime. What happens when we compare them to the inertial regime with rigid interest rates? In that regime, the sign of the macroprudential policy can be derived analytically, extending the first part of Proposition 3. Proposition 4. With CRRA preferences and δ 1 = 0, in the inertial regime there is always overborrowing d dd E V A D, D, r, s) + βv B D, D, r, s) < 0. Figure 2 illustrates this result using our numerical example. The figure is constructed as Figure 1, but choosing the parameters α 0, α 1 so as to go through the output targeting regime and the inertial regime with an optimal macroprudential tax. As the figure shows, in this region of the policy space monetary policy and macroprudential policy are substitutes. That is, along the line that connects the two regimes considered, the cross derivative 2 W/ r B τ is positive. As we compare the inertial regime to the output targeting regime, a more aggressive reduction of the interest rate in the second regime leads to a lower macroprudential tax. To complete the analysis of our example, it remains to compare the two extreme regimes, the inertial regime and the proactive regime. We know that in both regimes the borrowing tax is positive. But in what regime is the tax larger? And in what regime 19

15 10 Inertial regime Borrowing tax %) 5 0-5 Output targeting regime -10-15 5 10 15 20 25 Interest rate in the bad state %) Figure 2: Monetary policy, macroprudential tax and welfare: substitutability are the welfare benefits of the tax larger? In Table 2 we answer this questions in the context of our example. The table reports the value of the optimal tax, the difference between D at τ = 0 and D at the optimal tax, and the welfare gain due to the introduction of the optimal borrowing tax, in equivalent consumption terms. In the inertial regime all these measures of overborrowing are larger than in the proactive regime. Therefore, over the range of monetary policies that connect the two regimes, the forces of substitutability dominate the forces of complementariety identified in the previous subsection, so a more interventionist monetary policy ends up requiring less macroprudential intervention. The table also shows how the results change when we choose a lower value of γ. All measures of overborrowing are larger in this case, but the relative comparison of the two regimes is qualitatively the same. In the appendix, we experiment with different parameter combinations and our qualitative results do not seem to depend on the choice of β, γ, ψ, and φ. We can then give a more nuanced answer the question: does a more aggressive monetary regime increase or decrease the need for ex ante regulation? The answer turns out to be sensitive not only to the parameters of the model, but to the points on the policy space we are analyzing. If we compare a monetary policy that does not respond at all to negative shocks to a more aggressive policy, the need for macroprudential policy is lower, as the more aggressive monetary policy mitigates the aggregate demand externality. If we compare a monetary policy that is already responding strongly enough so as to elimi- 20

Table 2: Overborrowing in the inertial and proactive regimes Value of γ 0.5 0.3 Monetary regime Inertial Proactive Inertial Proactive Optimal borrowing tax 5.2% 0.2% 8.1% 1.0% Reduction in D due to the tax 0.025 0.0010 0.019 0.003 Welfare gain of tax in %) 0.014 0.00002 0.023 0.0005 nate the output gap to an even more aggressive policy, then the need for macroprudential policy increases in the latter regime. The conclusions above are of course partly driven by the special nature of our example. In all the numerical examples we have explored the benefit of macroprudential policy are lower in the proactive regime as compared to the inertial regime, but we do not know if that can be proved in general. The general message we can draw from these examples is that moral hazard concerns should not be necessarily stronger in presence of a more aggressive monetary policy, and that whether macroprudential policy needs to be tighter or looser depends on where we are in the policy space. 5.4 Another example of substitutability The examples considered so far show that once the central bank goes over and above its output stability objective to achieve a financial stability objective, a corrective macroprudential tax ex ante is welfare improving. One may wonder whether this is a general result. We now look at an example that shows that it is not. We maintain our benchmark parametrization but no longer assume zero dividends in period 1. Instead we assume that the dividends in the two periods are perfectly collinear δ 1 = ξδ 2, where ξ 0. We continue to assume δ 2 has a binary distribution with values {1.05, 1.5} and equal probabilities. Notice that borrowers wealth in period 1 is δ 1 + δ 2 /1 + r) D. Increasing ξ tilts the dividend shocks towards period 1. This makes monetary policy less powerful at buffering wealth shocks. We will see shortly how this affects optimal policy. Figure 3 shows the optimal macroprudential tax in the proactive regime for different levels of ξ. The output targeting regime features a zero tax, as in all previous examples. When ξ = 0 we replicate the result in the proactive column of Table 2: the optimal tax is positive and is 0.2%. However, as ξ increases the optimal tax is lower and when ξ is high enough the optimal tax becomes a subsidy, so we have underborrowing in the proactive 21

0.25 0.2 Borrowing tax %) 0.15 0.1 0.05 0-0.05 0 0.5 1 1.5 2 ξ Figure 3: Optimal macroprudential tax with positive dividends in period 1 regime. Recall the intuition for the complementarity result in Subsection 5.2. Optimal monetary policy generates a negative output gap in the bad state and a positive output gap in the good state. Increasing debt ex ante improves welfare in the bad state and reduces welfare in the good state. In the example of Subsection 5.2 the latter effect was larger. In the example here, with ξ high enough, the first effect is larger, so subsidizing borrowing is welfare improving. To get some intuition for why the magnitudes of the two effects are flipped with high ξ, notice that when δ 1 increases relative to δ 2 it gets harder for monetary policy to provide insurance to the B agents, as their wealth becomes less sensitive to r. This leads monetary policy in the bad state to lower rates more and produce a larger negative output gap. 6 Unconventional monetary policy So far we have focused on a monetary policy response that uses the traditional tool of setting nominal interest rates. Recent experience has seen central banks use nonconventional tools, such as asset purchases, to fight recessions, after nominal rates hit the zero lower bound. In this section, we extend the model to ask how the use of nonconventional tools affects the message of the previous sections. We consider here a simple variation on the case of log preferences and zero dividends at t = 1 studied in Section 4. As argued at the end of that section, it is easy to add a ZLB constraint to the policy prob- 22

lem in the form r r. We will do so and explore what happens when the ZLB is binding and the central bank is considering unconventional interventions. Our question is now: do ex post unconventional interventions that prop up asset prices increase the need for ex ante macroprudential policy? Before addressing that question, we need to characterize the equilibrium and introduce unconventional monetary policy. 6.1 Equilibrium with collateral constraints Assume B agents are not allowed to pledge the entire future value of the asset when they borrow in period t = 1, but only a fraction θ of it. This introduces a form of market segmentation that will make unconventional interventions have some bite. Then we need to rewrite the budget constraints of the B agents in periods 1 and 2 as follows xq + c B 1 Q D + d 1, c B 2 δ 2x 1 + r) d 1, where Q denotes the asset price, x is the amount of the asset kept by B agents at the end of period 1, and d 1 is borrowing in period 1. The collateral constraint is 1 + r) d 1 θδ 2 x. Here we allow the agent to sell some of the asset and keep x of it. We introduce this possibility because we want to model unconventional monetary policy as asset purchases by the central bank. In particular, we assume that the central bank raises a lump sum tax T on A agents and uses it to purchase x G units of the asset, so T = x G Q. The asset then produces goods in the government sector according to the strictly concave production function F x G), and these goods are paid to the A agents. The intertemporal constraint of the A agents then takes the form c A 1 + 1 1 + r ca 2 na 1 + D T + 1 1 + r F x G). 13) 23

We assume F 0) < δ 2 for all realizations of δ 2, so it will be inefficient to purchase the asset in normal circumstances. This policy features a tradeoff between increased purchases, which push up the asset price and improve the balance sheet of B agents, and the inefficiency cost of the government technology. 10 We now want to characterize the optimization problem of the B agents, imposing equilibrium in the asset market for a given value of government purchases x G and a given interest rate r. Two cases are possible. First, if the collateral constraint is not binding then the asset price is and the consumption levels are Q = δ 2 1 + r, c1 B = 1 ) δ2 1 + β 1 + r D, c2 B = β 1 + r) 1 + β ) δ2 1 + r D, exactly as in Section 4. After some algebra, the condition that ensures that the collateral constraint is not binding can be written as ) β δ2 1 + β 1 + r D 1 θ) x δ 2 1 + r. 14) So, as long as the B agents are solvent, meaning that their assets valued at the first best price are sufficient to repay their debt D, there is always a value of x low enough that ensures an unconstrained equilibrium. That is, given that x G = 1 x, if the government purchases enough of the asset at the first best price, the B agents are unconstrained. The second case is when the collateral constraint is binding, which arises when inequality 14) is violated. With a binding collateral constraint the values of c B 1, cb 2, and Q must satisfy: c B 1 = 1 x) Q + θδ 2x 1 + r D, c2 B = 1 θ) δ 2x, 15) ) 1 Q = θ 1 + r δ 2 + 1 θ) β u c B 2 u c1 B )δ 2, 16) ) ) u c1 B > β 1 + r) u c2 B. 10 This is different from a policy that subsidizes directly the asset price. If such policy could be financed by lump sum taxes on A agents, it would effectively transfer resources between the agents with no efficiency cost, thus allowing the planner to reach the first best. 24